Evaluation of Methods for the Calculation of the pKa of Cysteine Residues in Proteins

Methods for the calculation of the pK(a) ionizable amino acids are valuable tools for understanding pH-dependent properties of proteins. Cysteine is unique among the amino acids because of the chemical reactivity of its thiol group (S-H), which plays an instrumental role in several biochemical and regulatory functions. The acidity of noncatalytic cysteine residues is a factor in their susceptibility to chemical modification. Despite the plethora of existing pK(a) computing methods, no definitive protocol exists for accurately calculating the pK(a)'s of cysteine residues in proteins. A cysteine pK(a) test set was developed, which is comprised of 18 cysteine residues in 12 proteins where the pK(a)'s have been determined experimentally and an experimental structure is available. The pK(a)'s of these residues were calculated using three methods that use an implicit solvent model (H++, MCCE, and PROPKA) and an all-atom replica-exchange thermodynamic integration approach with the CHARMM36 and AMBER ff99SB-ILDNP force fields. The models that use implicit solvation methods were generally unreliable in predicting cysteine residue pK(a)'s, with RMSDs between 3.41 and 4.72 pK(a) units. On average, the explicit solvent methods performed better than the implicit solvent methods. RMSD values of 2.40 and 3.20 were obtained for simulations with the CHARMM36 and AMBER ff99SB-ILDNP force fields, respectively. Further development of these methods is necessary because the performance of the best method is similar to that of the null-model (RMSD = 2.74) and these differences in RMSD are of limited statistical significance given the small size of our test set.